[
https://issues.apache.org/jira/browse/FLINK-5176?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16243712#comment-16243712
]
Till Rohrmann commented on FLINK-5176:
--------------------------------------
This is not done completely for all components. E.g. {{TaskExecutor}} in
{{TaskManagerRunner}}, {{ResourceManager}}, {{Dispatcher}} in
{{SessionClusterEntrypoint}} and {{JobmanagerRunner}} in
{{JobClusterEntrypoint}}.
> Properly wait for component shutdown in Runner components
> ---------------------------------------------------------
>
> Key: FLINK-5176
> URL: https://issues.apache.org/jira/browse/FLINK-5176
> Project: Flink
> Issue Type: Improvement
> Components: Distributed Coordination
> Reporter: Till Rohrmann
> Assignee: Shuyi Chen
> Priority: Minor
> Labels: flip-6
>
> Currently, the different runners for the Flip-6 components {{JobMaster}},
> {{ResourceManager}} and {{TaskExecutor}} don't wait until the component has
> been shut down in their own {{shutDown}} method. The problem is that the
> {{AkkaRpcService}} implements the {{RpcService#shutDown}} method via a
> {{PoisonPill}} which is not executed synchronously. Instead we should obtain
> the {{RpcEndpoint#getTerminationFutures}} and wait on the returned future
> (with a given timeout).
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
